CONCLUSIONES

Once the needs have been defined in the logical network design, the next step is to develop a physical network design (or set of possible designs). The physical network design starts with the client and server computers needed to support the users and applications. If the network is a new network, new computers will need to be purchased. If the network is an existing network, the servers may need to be upgraded to the newest technology. Once these are designed, then the circuits and devices connecting them are designed.

6.3.1 Designing Clients and Servers

The idea behind the building-block approach is to specify needs in terms of some standard units. Typical users are allocated the base-level client computers, as are servers supporting typical applications. Users and servers for applications needing more powerful computers are assigned some advanced computer. As the specifications for computers rapidly improve and costs drop (usually every 6 months), today’s typical user may receive the type of com-puter originally intended for the advanced user when the network is actually implemented, and the advanced users may end up with a computer not available when the network was designed.

6.3.2 Designing Circuits

The same is true for network circuits and devices (e.g., hubs, routers, switches). There are two interrelated decisions in designing network circuits and devices: the fundamental technology and protocols (e.g., Ethernet) and the capacity of each circuit (e.g., 100 Mbps, 1000 Mbps). These are interrelated because each technology offers different circuit capacities.

Designing the circuit capacity means capacity planning, estimating the size and type of the standard and advanced network circuits for each type of network (LAN, backbone, WAN). As you will learn in Chapter 7 on LANs, wired and wireless circuits come in standard sizes. Most users with a desktop or laptop computer don’t need to send files that are over a gigabyte in size at a time (i.e., 1000 Meg). And if they do, they understand there may be a slight delay. Therefore, circuits for wired LANs are typically 100 Mbps or 1 Gbps. Wireless circuits are a little different, so we’ll avoid them until Chapter 7.

Designing circuit capacities for backbone networks is more challenging because back-bones move traffic from many computers at one time and there are more choices in standard sizes. This requires some assessment of the current and future circuit loading (the amount of data transmitted on a circuit). This analysis can focus on either the average circuit traf-fic or the peak circuit traftraf-fic. For example, in an online banking network, traftraf-fic volume peaks usually are in the midmorning (bank opening) and just prior to closing. Airline and rental car reservations network designers look for peak volumes before and during holidays or other vacation periods, whereas telephone companies normally have their highest peak volumes on Mother’s Day. Designing for peak circuit traffic is the ideal.

The designer usually starts with the total characters transmitted per day on each cir-cuit or, if possible, the maximum number of characters transmitted per 2-second interval if peaks must be met. You can calculate message volumes by counting messages in a current network and applying some estimated growth rate. If an existing network is in place, net-work monitors/analyzers (see Chapter 12) may be able to provide an actual circuit character count of the volume transmitted per minute or per day.

A good rule of thumb is that 80% of this circuit loading information is easy to gather.

The last 20% needed for very precise estimates is extremely difficult and expensive to find.

However, precision usually is not a major concern because of the stairstep nature of commu-nication circuits and the need to project future needs. For example, the difference between 100 Mbps and 1Gbps is quite large, and assessing which level is needed for typical traffic

does not require a lot of precision. Forecasts are inherently less precise than understanding current network traffic. The turnpike effect is an expression that means that traffic increases much faster than originally forecast. It comes from the traffic forecasting that was done for the construction of the early interstate highways. When a new, faster highway (or network) is built, people are more likely to use it than the old slow one because it is available, is very efficient, and provides new capabilities. The annual growth factor for network use may vary from 5% to 50% and, in some cases, may exceed 100% for high-growth organizations.

Although no organization wants to overbuild its network and pay for more capacity than it needs, in most cases, upgrading a network costs 50% to 80% more than building it right the first time. Few organizations complain about having too much network capacity, but being under capacity can cause significant problems. Given the rapid growth in network demand and the difficulty in accurately predicting it, most organizations intentionally over-build (over-build more capacity into their network than they plan to use), and most end up using this supposedly unneeded capacity within 3 years.

In any network, there may be a bottleneck, a circuit that is filled almost to its capacity and thus is the critical point that determines whether users get good or bad response times.

When users complain about a slow network, it is usually because there is a bottleneck circuit somewhere in the network. Of course, the bottleneck could also be a slow Web server that is simply receiving more traffic than it can handle, but usually the problem is a circuit.

Take another look at Figure 6-4. Suppose we specified 1 Gbps circuits as the standard for the LANs. If each LAN has 20 computers, then this is in theory a total capacity of 120 Gbps in the building (6 LANs × 20 computers each × 1 Gbps = 120 Gbps). Not all the computers will be sending or receiving at the same time, so this is artificially high, but it is a theoretical maximum.

If this is the case, what speed should we specify for the building backbone? We have a few standard speeds, as you will learn in Chapter 8: 1 Gbps, 10 Gbps, 40 Gbps, 100 Gbps.

A 1 Gbps backbone is probably too slow and would end up being a bottleneck. Is 10 Gbps enough? It’s hard to say without knowing the circuit loading. Without the circuit loading, most network designers would set the building backbone speed at one level above the stan-dard LAN speed, which in this case would be 10 Gbps.

This problem continues at the next architecture component—the campus core back-bone. If each building has a 10 Gbps backbone, what speed should the campus backbone that connects all the buildings be? Without a circuit loading, it’s hard to say. Once again, most network designers would set the building backbone speed at one level above the build-ing backbone speed, which in this case would be 40 Gbps. And this is where reality sets in.

Today, the technology for 40 Gbps is very expensive—so expensive, in fact, that most orga-nizations don’t buy it unless they really need it. Chances are, the campus backbone would be designed at 10 Gbps, which means it might be the bottleneck—at least for traffic on campus.

Figure 6-5 shows the physical design for the network in Figure 6-4. Take a moment to look at it and compare Figures 6-4 and 6-5.

As we move beyond the campus to the enterprise edge, network design becomes a bit more difficult. As you will learn in Chapter 9, on WANs, and Chapter 10, on the Internet, the technologies we use for WANs and Internet access are quite different to what we use for LANs and backbones. Their speeds are much, much slower and much more expensive.

A typical WAN circuit speed is between 1 Mbps and 50 Mbps. Yes, that was Mbps; in other words, more than 100 times slower than the speed of our backbone networks. Thus the bottleneck in most enterprise networks is the WAN and the Internet, not the enterprise campus network.

This is also true for the network in your house or apartment. Most wireless LAN access points you buy today provide speeds of 100–400 Mbps, yet your Internet connection is usu-ally less than 25 Mbps. This means the response times you experience when you’re on the Internet will be the same whether you buy a really fast, state-of-the-art wireless access point

Technology Design 177

FIGURE 6-5

Physical network design for a single building _Floor^3rd

2nd Floor

1st Floor

1 West

3 East 3 West

2 East

2 West

1 East

Router switch switch switch

switch

switch

switch switch

Campus Core Backbone 1 Gbps switch

10 Gbps switch 10 Gbps router

or an old one that provides only 50 Mbps, because the bottleneck is the Internet access, not the wireless LAN. Unless you’re spending a lot of money on a really fast Internet connection, don’t waste your money on a really fast wireless access point.

6.3.3 Network Design Tools

Network modeling and design tools can perform a number of functions to help in the tech-nology design process. With most tools, the first step is to enter a diagram or model of the existing network or proposed network design. Some modeling tools require the user to create the network diagram from scratch. That is, the user must enter all of the network components by hand, placing each server, client computer, and circuit on the diagram and defining what each is.

Other tools can “discover” the existing network; that is, once installed on the network, they will explore the network to draw a network diagram. In this case, the user provides some starting point, and the modeling software explores the network and automatically draws the diagram itself. Once the diagram is complete, the user can then change it to reflect the new network design. Obviously, a tool that can perform network discovery by itself is most helpful when the network being designed is an upgrade to an existing network and when the network is very complex.

Once the diagram is complete, the next step is to add information about the expected network traffic and see if the network can support the level of traffic that is expected. Simu-lation, a mathematical technique in which the network comes to life and behaves as it would under real conditions, is used to model the behavior of the communication network. Appli-cations and users generate and respond to messages while the simulator tracks the number of packets in the network and the delays encountered at each point in the network.

Simulation models may be tailored to the users’ needs by entering parameter values specific to the network at hand (e.g., this computer will send an average of three 100-byte

packets per minute and receive one hundred 1,500-byte packets per minute). Alternatively, the user may prefer to rely primarily on the set of average values provided by the network.

Once the simulation is complete, the user can examine the results to see the estimated response times throughout. It is important to note that these network design tools provide only estimates, which may vary from the actual results. At this point, the user can change the network design in an attempt to eliminate bottlenecks and rerun the simulation. Good modeling tools not only produce simulation results but also highlight potential trouble spots (e.g., servers, circuits, or devices that experienced long response times). The very best tools offer suggestions on how to overcome the problems that the simulation identified.

6.3.4 Deliverables

The key deliverable is a set of one or more physical network designs like that in Figure 6-5, which is the design for a single building. Most designers like to prepare several physical designs so they can trade-off technical benefits (e.g., performance) against cost. In most cases, the critical part is the design of the network circuits and devices. In the case of a new network designed from scratch, it is also important to define the client computers with care because these will form a large portion of the total cost of the network. Usually, however, the network will replace an existing network and only a few of the client computers in the existing network will be upgraded.